This is an example of an ecologic design. Four counties with variable rates of air pollution had the number of children with prescriptions of various antipsychotic or sedative medications compared. There is (for the TL:DR group a statistically significant but weak correlation).
The medication group used as outcome in this study includes any medication with a Swedish ATC code starting with ‘N05’, hereafter referred to as N05. N05 consists of neuroleptics (antipsychotic medications), ataractics and sleeping pills (a broad group of sedative medications including hydroxyzine and melatonin-based medications). The majority of dispensed N05 medications for children and adolescents are sedative medications and sleeping pills. It should be noted that antidepressants or attention deficit hyperactivity disorder (ADHD) medications are not included in the N05 group. We had access to the number of times each year during follow-up an individual had dispensed an N05 medication (figure 2), but for integrity reasons, we could not obtain data on the exact type of medication within the N05 group. We defined an event as a dispensed N05 medication at least once during follow-up. Individuals who had a record of an N05 dispensed medication from July 2005 (start date of the register) to 31 December 2006 (the year and a half before start of follow-up) were excluded from the analysis. An event was recorded the first year during follow-up that a N05 medication was dispensed and the event date was set to 1 July (we did not have access to the date each respective year the medication was dispensed). Each cohort member was followed up until an event occurred, loss-to-follow-up (meaning that they had no longer a registered address in the study area), death, 18th birthday or end of follow-up, whichever came first. We used an open cohort approach and continuously included cohort members from date of birth, or individuals moving to the study area from 1 January the year after they moved in. For practical reasons, an individual who moved out of the study area was censored and not included again if he or she moved back. We excluded individuals whose parents had N05 medications dispensed since the start date of the register. The total size of the cohort after exclusions was 896?117 individuals.
This is from the results section. HR means Hazard Ratio. The number of events is the number of prescriptions (per 1000) in various regions. The numbers are common in these studies, where national datasets are used.
The HRs regarding PM10 and PM2.5 were very similar. We therefore present only the results for PM10 and NO2. There were in total 18?675 individuals who at least once had dispensed N05 during follow-up, with a total number of person-years followed up of 3?101?756 (table 1). The average length of follow-up was 3.5?years. The number of events per 1000 cohort members was 23 in Västra Götaland, 23 in Västerbotten, 20 in Stockholm and 20 in Skåne (table 1). The agreement between an event in 1?year and an event in subsequent years was not very high; for example, the ? statistic calculated for an event in 2007 and 2008 was 0.21, in 2007 and 2009 was 0.14, and in 2007 and 2010 was 0.11. The highest concentrations of NO2 were found in Stockholm (yearly median: 8.3?µg/m3) and the highest levels of PM10 were found in Skåne (yearly mean median: 15.8?µg/m3; table 1). The correlation between NO2 and PM10 was, due to the modelling of PM10, high, and ranged between 0.83 (Västerbotten), 0.97 (Skåne) and 0.98 (Stockholm and Västra Götaland). ….
The HR in association with a 10?µg/m3 increase in NO2 adjusted for age at the start of follow-up, sex, maternal and paternal education, maternal body mass index in early pregnancy, maternal smoking during early pregnancy and group-level education levels was 1.09 (95% CI 1.06 to 1.12; table 3). The corresponding HR associated with a 10?µg/m3 increase in PM10 was 1.04 (95% CI 1.00 to 1.08). The correlation between NO2 and PM10 was too high for them to be included simultaneously in the model, because NO2 was one of the predictors of urban background PM10 in the PM10 model. When using the backward selection technique to identify what variables should be included in the model, all variables remained in the model due to their low p values, except the group-level SES variable that had a p value >0.20. The estimates with and without that variable included were very similar (results not shown). The crude HRs were close to one (no association), and the association appeared when adjusting for age. The other variables did not seem to have any substantial influence on the estimates (results not shown). The associations seemed heterogeneous across the four counties (table 3 and figure 4A,B), at least for the association with PM10 (p=0.001). For NO2, the p value for effect modification was 0.24. For example, the association with a 10?µg/m3 increase in NO2 was quite similar in Stockholm, Västra Götaland and Västerbotten (HRs of 1.13, 1.11 and 1.13, respectively), whereas there was no evident association (HR=1.03) in Skåne (table 3 and figure 4A,B). The HRs associated with a 10?µg/m3 increase in PM10 were quite similar in Stockholm and Västra Götaland (1.23 and 1.18, respectively), 1.00 in Skåne and 1.89 in Västerbotten. The high estimate in Västerbotten is uncertain, a contributing factor that may be low statistical power to detect differences. The estimates seemed similar among boys and girls (table 3), and the p value for effect modification was 0.30 for NO2 and 0.82 for PM10
And this is when the results become a little spurious. The effect size is probably around 1.5: that is a correlation: it is not causation. The authors note that it is only when they correct for factors that a difference emerges.
At best this indicates there may be a hypothesis worth investigating m– for even the highest level of pollution would be considered “safe air” by the EU. There are some virtues: this work is cheap, no one is harmed, and Scandinavia has a tradition of using public databases for epidemiology.
Ecologic studies ask questions. They generate models. They prove or disprove not a thing.